Advanced segmentation method for integrating multi-omics data for early cancer detection

Abstract

The global burden of cancer underscores the critical need for early diagnosis. Traditional diagnostic methods, relying on single biomarkers or imaging, often lack comprehensive predictive accuracy. Existing systems often focus on one or two types of omics data, such as genome or transcriptome, but do not comprehensively integrate multiple omics layers (genomic, transcriptomic, proteomic, and epigenomic). This limitation restricts the ability to capture the full biological complexity and heterogeneity of cancer, which can be critical for accurate prediction and understanding of disease mechanisms. We propose an advanced cancer prediction method called Integrated Multi-Omics Segmentation (IMOS), which enhances the processing of multi-omics data by integrating genomic, transcriptomic, proteomic, and epigenomic information. IMOS segments data into biologically meaningful regions, facilitating more precise analysis. IMOS achieves outstanding performance with an average precision of 92 %, sensitivity of 88 %, and specificity of 94 %, outperforming traditional methods by 15 % in precision, 10 % in sensitivity, and 8 % in specificity. Validation using the Genomic Data Commons (GDC) dataset, encompassing diverse cancer types, demonstrated IMOS’s robustness with accuracy of 91 %, sensitivity of 87 %, and specificity of 93 %. The system also excels in clustering evaluation, with a silhouette score ranging from 0.55 to 0.62 and the lowest Davies-Bouldin index achieved with three clusters.